Abstract

Sediment transport has attracted the attention of engineers from various aspects and different methods have been used for its estimation. So, several experimental equations have been submitted by experts. Though the results of these methods have considerable differences with each other and with experimental observations, because the sediment measures have some limits, these equations can be used in estimating sediment load. With regard to the fact that Givichay River has high sediment production in the region, it is chosen as the study area. This river is one of the Qezeuzan River branches and through this river it joins to the Sefidrud River. Sefidrud dam is one of the most sediment receiver dams in the world and now more than half of the dam capacity has been filled with sediment. With regard to the fact that there are not enough sediment measure stations in this region, therefore different methods have been used. In this study, neural differential evolution (NDE) models are proposed to estimate suspended sediment concentration. NDE models are improved by combining two methods, neural networks and differential evolution. In the first part of the study, NDE model is trained using daily river flow and suspended sediment data belonging to Givi Chay River in northwest of Iran and various combinations of current daily stream flow and past daily stream flow, suspended sediment data are used as inputs to the NDE model so as to estimate current suspended sediment. In the second part of the study, the suspended sediment estimations provided by NDE model are compared with multi layer perceptron (MLP), radial basis function (RBF) and sediment rating curves (SRC) results. The Root mean squared error (RMSE) and the determination coefficient (R²) are used as comparison criteria. Obtained results demonstrate that NDE are in good agreement with the observed suspended sediment concentration; while they depict better results than RBF, MLP and SRC methods. For example, in Givi Chay River station, the determination coefficient (R²) is 0.9621 for NDE model, while it is 0.9114, 0.90 and 0.8447 for RBF, MLP and SRC models, respectively. However, for the estimation of maximum sediment peak, the NDE was mostly found to be better than the RBF and the other techniques. The results also indicate that the NDE may provide better performance than the RBF, MLP and SRC in the estimation of the total sediment load (Re = -26%).

Key words: Givi Chay River, neural differential evolution, multi-layer perceptron model,radial basis function, sediment rating curves.